Linear Unmixing Model Research Articles

Hyperspectral Blind Unmixing Using a Double Deep Image Prior.

With the rise of machine learning, hyperspectral image (HSI) unmixing problems have been tackled using learning-based methods. However, physically meaningful unmixing results are not guaranteed without proper guidance. In this work, we propose an unsupervised framework inspired by deep image prior (DIP) that can be used for both linear and nonlinear blind unmixing models. The framework consists of three modules: 1) an Endmember estimation module using DIP (EDIP); 2) an Abundance estimation module using DIP (ADIP); and 3) a mixing module (MM). EDIP and ADIP modules generate endmembers and abundances, respectively, while MM produces a reconstruction of the HSI observations based on the postulated unmixing model. We introduce a composite loss function that applies to both linear and nonlinear unmixing models to generate meaningful unmixing results. In addition, we propose an adaptive loss weight strategy for better unmixing results in nonlinear mixing scenarios. The proposed methods outperform state-of-the-art unmixing algorithms in extensive experiments conducted on both synthetic and real datasets.

Retrieval of Fractional Vegetation Cover from Remote Sensing Image of Unmanned Aerial Vehicle Based on Mixed Pixel Decomposition Method

FVC (fractional vegetation cover) is highly correlated with wheat plant density in the reviving period, which is an important indicator for conducting variable-rate nitrogenous topdressing. In this study, with the objective of improving inversion accuracy of wheat plant density, an innovative approach of retrieval of FVC values from remote sensing images of a UAV (unmanned aerial vehicle) was proposed based on the mixed pixel decomposition method. Firstly, remote sensing images of an experimental wheat field were acquired by using a DJI Mini UAV and endmembers in the image were identified. Subsequently, a linear unmixing model was used to subdivide mixed pixels into components of vegetation and soil, and an abundance map of vegetation was acquired. Based on the abundance map of vegetation, FVC was calculated. Consequently, a linear regression model between the ground truth data of wheat plant density and FVC was established. The coefficient of determination (R2), RMSE (root mean square error), and RRMSE (Relative-RMSE) of the inversion model were calculated as 0.97, 1.86 plants/m2, and 0.677%, which indicates strong correlation between the FVC of mixed pixel decomposition method and wheat plant density. Therefore, we can conclude that the mixed pixel decomposition model of the remote sensing image of a UAV significantly improved the inversion accuracy of wheat plant density from FVC values, which provides method support and basic data for variable-rate nitrogenous fertilization in the wheat reviving period in the manner of precision agriculture.

End-to-End Convolutional Autoencoder for Nonlinear Hyperspectral Unmixing

Hyperspectral Unmixing is the process of decomposing a mixed pixel into its pure materials (endmembers) and estimating their corresponding proportions (abundances). Although linear unmixing models are more common due to their simplicity and flexibility, they suffer from many limitations in real world scenes where interactions between pure materials exist, which paved the way for nonlinear methods to emerge. However, existing methods for nonlinear unmixing require prior knowledge or an assumption about the type of nonlinearity, which can affect the results. This paper introduces a nonlinear method with a novel deep convolutional autoencoder for blind unmixing. The proposed framework consists of a deep encoder of successive small size convolutional filters along with max pooling layers, and a decoder composed of successive 2D and 1D convolutional filters. The output of the decoder is formed of a linear part and an additive non-linear one. The network is trained using the mean squared error loss function. Several experiments were conducted to evaluate the performance of the proposed method using synthetic and real airborne data. Results show a better performance in terms of abundance and endmembers estimation compared to several existing methods.

Robust linear unmixing with enhanced constraint of classification for hyperspectral remote sensing imagery

Although hyperspectral data, especially spaceborne images, are rich in spectral information, their spatial resolution is usually low due to the limitation of sensor design and other factors. Therefore, for the application of hyperspectral images, unmixing technology is a key processing technology, such as linear mixing model and its derived algorithms have made a certain progress. However, a real scene often contains both pure and mixed pixels. The existing methods usually ignore the consideration and analysis of this situation in the process of model design and simulation experiment. In this context, this paper proposes a robust linear unmixing model with the enhanced constraint of classification for hyperspectral image. In general, it designs a framework combining unmixing and classification. In the task for real scene data, endmembers are extracted first, and then the hard classification term constructed after the expansion of endmembers (training samples) based on similarity is introduced to provide the sparsity constraint of the overall model, so as to realize relatively complete adjustment and effective image unmixing under complex conditions. Considering the scene with different distributions, the simulation experiment designs several groups of data tests, including different proportions of pure and mixing pixels. The unmixing results of three simulated datasets and two real datasets show that the unmixing results of this method are better than those of the other six comparison methods. This model improves the accuracy of unmixing and realizes effective unmixing.

Hyperspectral linear unmixing based on collaborative sparsity and multi-band non-local total variation

ABSTRACT In recent years, spectral unmixing is a hot issue in the hyperspectral remote sensing technology and applications. Within various methods, sparse regression is one of the most commonly used methods in the hyperspectral unmixing community. In order to overcome the limitation of spatial correlation of existing sparse unmixing methods and improve further the accuracy of sparsity representation, a novel hyperspectral linear unmixing method via fusion of collaborative sparsity and multi-band non-local total variation is proposed. This method analyses the sparsity and spatially geometrical structure of hyperspectral pixels by studying the linear spectral mixed model of hyperspectral images, and reveals the interior spatial correlation of hyperspectral images with the non-local variation framework. Specifically, the sparsity and spatial correlation of hyperspectral imagery are discussed, and a hyperspectral linear sparse unmixing model is afterwards constructed by combining the relaxation technique and non-local total variation framework. Second, the solution procedure of the above spectral unmixing model is divided into two steps under the variable separation method: fractional abundance estimation and abundance map restoration. The alternating direction method of multipliers (ADMM) and split Bregman operator are exploited to obtain the spectral unmixing results. Finally, the proposed hyperspectral unmixing algorithm is evaluated with synthetic and real hyperspectral datasets. In the experiments with a synthetic hyperspectral dataset, the feasibility and effectiveness of the method are analysed quantitatively and qualitatively. The quantitative metrics and visual examination of the estimated fractional abundance map are also better than the performance of the current mainstream hyperspectral sparse unmixing algorithms. Furthermore, two real hyperspectral datasets are applied to the algorithm of this work to prove its practicability.

An Adaptive-Parameter Pixel Unmixing Method for Mapping Evergreen Forest Fractions Based on Time-Series NDVI: A Case Study of Southern China

Spectral unmixing remains the most popular method for estimating the composition of mixed pixels. However, the spectral-based unmixing method cannot easily distinguish vegetation with similar spectral characteristics (e.g., different forest tree species). Furthermore, in large areas with significant heterogeneity, extracting a large number of pure endmember samples is challenging. Here, we implement a fractional evergreen forest cover-self-adaptive parameter (FEVC-SAP) approach to measure FEVC at the regional scale from continuous intra-year time-series normalized difference vegetation index (NDVI) values derived from moderate resolution imaging spectroradiometer (MODIS) imagery acquired over southern China, an area with a complex mixture of temperate, subtropical, and tropical climates containing evergreen and deciduous forests. Considering the cover of evergreen forest as a fraction of total forest (evergreen forest plus non-evergreen forest), the dimidiate pixel model combined with an index of evergreen forest phenological characteristics (NDVIann-min: intra-annual minimum NDVI value) was used to distinguish between evergreen and non-evergreen forests within a pixel. Due to spatial heterogeneity, the optimal model parameters differ among regions. By dividing the study area into grids, our method converts image spectral information into gray level information and uses the Otsu threshold segmentation method to simulate the appropriate parameters for each grid for adaptive acquisition of FEVC parameters. Mapping accuracy was assessed at the pixel and sub-pixel scales. At the pixel scale, a confusion matrix was constructed with higher overall accuracy (87.5%) of evergreen forest classification than existing land cover products, including GLC 30 and MOD12. At the sub-pixel scale, a strong linear correlation was found between the cover fraction predicted by our method and the reference cover fraction obtained from GF-1 images (R2 = 0.86). Compared to other methods, the FEVC-SAP had a lower estimation deviation (root mean square error = 8.6%). Moreover, the proposed method had greater estimation accuracy in densely than sparsely forested areas. Our results highlight the utility of the adaptive-parameter linear unmixing model for quantitative evaluation of the coverage of evergreen forest and other vegetation types at large scales.

Remote sensing and GIS for urbanization and flood risk assessment in Phnom Penh, Cambodia

This study performed flood risk assessment in urbanized areas using geospatial and remotely sensed data for 1990–2005–2020 periods using the linear unmixing model (LUM), random forests (RF), and support vector machines (SVM). The urban mapping results verified with reference data indicated close agreement, with the overall accuracies and Kappa coefficients higher than 88.9% and 0.78, respectively. A remarkable increase of 14.9% in urban area was observed for the period 2005–2020, compared to 7.2% during the period 1990–2005. The results of flood risk modeling revealed that RF produced slightly more accurate results than SVM. The flood risk areas aggregated with urban maps showed that the larger urban flood risk area was especially observed for the period 2005–2020. The urban areas of high/very high and medium flood risks calculated for 2020 were 19.7% and 20%, respectively. Such flood risk areas were also overlaid with the population density for urban planning and management.

A Combined Unmixing Framework for Impervious Surface Mapping on Medium-Resolution Images with Visible Shadows

Spectral unmixing methods with medium-resolution remote sensing images have become the main approach to mapping urban impervious-surface information. However, as more tall buildings appear, numerous visible shadows exist in medium-resolution images; these have usually been ignored in previous research, but they seriously affect accuracy. To solve this problem, we propose a combined unmixing framework to extract impervious surface in nonshadow and shadow areas, using linear and nonlinear unmixing models, respectively. First shadow is separated from nonshadow. Then a nonlinear unmixing method is selected to map impervious surface in shadow, which is more suitable to the complex imaging environment in shadow, and a classic linear unmixing model in nonshadow. Through experimental tests, the proposed combined unmixing framework is shown to effectively reduce error in two study areas compared with classical unmixing methods.

A novel inequality-constrained weighted linear mixture model for endmember variability

Spectral unmixing has trigged considerable attention to both multispectral and hyperspectral data processing. Among different kinds of spectral unmixing models, the linear mixture model (LMM) has been widely applied with its simplicity in light scattering mechanisms and flexibility in different scenarios. Recently, a variety of optimized LMMs have been developed to enhance the decomposing ability. However, two major challenges still exist: endmember variability and global optimum solution with the inequality constraint. In this study, aiming to avoid the influence of endmember variability and to extend the application of LMM, a novel inequality-constrained weighted linear mixture model (IWLMM) is proposed. Based on the errors-in-variables (EIV) model and the theory of weighted total least squares (WTLS), the issue of endmember variability and stochastic errors of both endmembers and mixed pixels are eliminated. With the non-negativity constraint, a brief optimal algorithm to solve the IWLMM is generated as well. In order to indicate the advantages of proposed IWLMM, comparative experiments with the conventional fully constrained least squares (FCLS), three currently proposed LMM-based models (perturbed LMM (PLMM), extended LMM (ELMM) and augmented LMM (ALMM)) and three physics-based nonlinear mixture models (bilinear-Fan model (BFM), generalized bilinear model (GBM) and polynomial post-nonlinear model (PPNM)) on four synthetic and real scenarios with hyperspectral and multispectral datasets were performed. The experimental results suggest the IWLMM outperforms the other seven linear and nonlinear unmixing models after adding the correction for endmember variability by modeling the perturbation and scaling factors of spectral features simultaneously. It has an impressive ability for decomposing and reconstructing pixels.

Hyperspectral Unmixing Based on Spectral and Sparse Deep Convolutional Neural Networks

Hyperspectral unmixing refers to the process of obtaining endmembers and abundance vectors through linear or nonlinear models. The traditional linear unmixing model assumes that each mixed pixel can be represented by a linear combination of endmembers. Considering real-world situations, a sparse constraint is normally added to the linear unmixing model. However, the linear model does not take into account that the spectrum of mixed pixels is not simply linearly mixed. To fully study the mixing characteristics of ground object spectra before being imaged by the sensor, we propose a supervised unmixing architecture based on one-dimensional convolutional neural network (CNN) by considering the spectral information and the sparse characteristics in the mixed pixel. Since one-dimensional CNN only consider feature learning, we combine the traditional root mean square error (RMSE) and l_"1" regularization in its loss function to minimize training error. The performance of our proposed unmixing model is assessed by comparing the unmixing results with three traditional linear sparse unmixing algorithms, and the Fuzzy-ARTMAP neural network in a simulated dataset and three real datasets. The RMSE was used to verify the unmixing accuracy of the different methods. The results showed that the RMSE obtained by our proposed CNN-based method was the lowest among the methods on all three real datasets, proving the effectiveness and stability of CNN in unmixing tasks.

Thermal emission measurements of ordinary chondrite mineral analogs in a simulated asteroid environment: 2. Representative mineral mixtures

We investigate a suite of synthetic mineral mixtures designed to act as bulk mineralogical analogs to H, L, and LL ordinary chondrite meteorites in order to probe how the thermal emission characteristics of such materials change between ambient and simulated asteroid environmental conditions. Due to the parent body link with certain S-type asteroids, studying these analog mixtures in an environment that is relevant to actual asteroid surfaces advances our understanding of the thermal emission properties of one of the most common regolith types among the main-belt and near-Earth asteroid populations. The observed changes in spectral emissivity features due to cold, vacuum conditions are not as large as previously observed for single mineral (silicate) samples. We interpret this difference to be the result of metallic and opaque components weakening near-surface thermal gradients in the mixtures. As such, we predict that near-surface thermal gradients on ordinary chondrite parent bodies (e.g., S-type asteroids) are likely much weaker than would be inferred from cold, vacuum measurements of individual mineral components. We tested whether the increased spectral contrast observed in fine-grained (<25 μm) samples measured in a cold, vacuum environment increases the efficacy of least squares linear unmixing methods. It is found that the accuracy of such models does not improve relative to measurements made under ambient conditions, thus linear unmixing models are not expected to yield accurate estimates of the modal mineralogy of airless planetary surfaces if they are dominated by fine-grained regolith. Mixtures with coarse particle sizes (125–250 μm) that were modeled using the coarse particle size endmembers yielded results that were largely independent of the environmental conditions, but with larger errors in spectral fits for samples measured in a simulated asteroid environment. At simulated asteroid environmental conditions, the bulk silicate composition and metal content play a more important role in determining the thermal state and brightness temperature of the sample than at ambient conditions. Modest changes in metal content (10–25 wt%) lead to large differences in the brightness temperature of a sample. Under simulated asteroid conditions, an ~10 K increase in maximum brightness temperature that tracks with increased iron content is observed at fine particle sizes (<25 μm) between each of the analog ordinary chondrite groups. Based on these results, it may be difficult to distinguish H, L, and LL compositions of suspected ordinary chondrite parent bodies using only Earth- or space-based thermal emission spectra. This is inferred from the spectral similarity of the analog mixtures and the absence of significant variation in spectral emissivity associated with reported differences in bulk metal content for ordinary chondrites.

An Overview on Linear Unmixing of Hyperspectral Data

Hyperspectral remote sensing technology has a strong capability for ground object detection due to the low spatial resolution of hyperspectral imaging spectrometers. A single pixel that leads to a hyperspectral remote sensing image usually contains more than one feature coverage type, resulting in a mixed pixel. The existence of a mixed pixel affects the accuracy of the ground object identification and classification and hinders the application and development of hyperspectral technology. For the problem of unmixing of mixed pixels in hyperspectral images (HSIs), the linear mixing model can model the mixed pixels well. Through the collation of nearly five years of the literature, this paper introduces the development status and problems of linear unmixing models from four aspects: geometric method, nonnegative matrix factorization (NMF), Bayesian method, and sparse unmixing.

Coupled Convolutional Neural Network With Adaptive Response Function Learning for Unsupervised Hyperspectral Super Resolution

Due to the limitations of hyperspectral imaging systems, hyperspectral imagery (HSI) often suffers from poor spatial resolution, thus hampering many applications of the imagery. Hyperspectral super-resolution refers to fusing HSI and MSI to generate an image with both high spatial and high spectral resolutions. Recently, several new methods have been proposed to solve this fusion problem, and most of these methods assume that the prior information of the Point Spread Function (PSF) and Spectral Response Function (SRF) are known. However, in practice, this information is often limited or unavailable. In this work, an unsupervised deep learning-based fusion method - HyCoNet - that can solve the problems in HSI-MSI fusion without the prior PSF and SRF information is proposed. HyCoNet consists of three coupled autoencoder nets in which the HSI and MSI are unmixed into endmembers and abundances based on the linear unmixing model. Two special convolutional layers are designed to act as a bridge that coordinates with the three autoencoder nets, and the PSF and SRF parameters are learned adaptively in the two convolution layers during the training process. Furthermore, driven by the joint loss function, the proposed method is straightforward and easily implemented in an end-to-end training manner. The experiments performed in the study demonstrate that the proposed method performs well and produces robust results for different datasets and arbitrary PSFs and SRFs.

Optical interpretation of oil emulsions in the ocean – Part II: Applications to multi-band coarse-resolution imagery

Water in oil (WO) and oil in water (OW) emulsions from marine oil spills have different physical properties, volume concentrations, and spectral characteristics. Identification and quantification of these different types of oil emulsions are important for oil spill response and post-spill assessment. While the spectral characteristics of WO and OW emulsions have been presented in previous studies including Part I of this series, their application to airborne and satellite imagery is further demonstrated here. Using AVIRIS and Landsat observations, we firstly show that false color Red-Green-Blue composite images from Landsat-like sensors (R: 1677 nm, G: 839 nm, B: 660 nm) are effective in differentiating WO and OW emulsions as they show reddish and greenish colors, respectively, in such composite images. This is a consequence of the relative difference in the reflectance of WO and OW emulsions at 1677 and 839 nm, which is not impacted by the presence of medium-strength sunglint or the surface heterogeneity within medium-resolution pixels (e.g., 30 m). Based on image statistics, a decision tree method is proposed to classify oil type, and oil quantification is further attempted, with results partially validated through spectral analysis and spatial coherence test. The numerical mixing experiments using AVIRIS pixels further indicate that the SWIR bands might be used to develop linear unmixing models in the future once the coarse-resolution oiled pixels are first classified to WO and OW types, and 1295 nm is the optimal wavelength to perform spectral unmixing of mixed coarse-resolution pixels.

Phenology-Based Mapping of an Alien Invasive Species Using Time Series of Multispectral Satellite Data: A Case-Study with Glossy Buckthorn in Québec, Canada

Glossy buckthorn (Frangula alnus Mill.) is an alien species in Canada that is invading many forested areas. Glossy buckthorn has impacts on the biodiversity and productivity of invaded forests. Currently, we do not know much about the species’ ecology and no thorough study of its distribution in temperate forests has been performed yet. As is often the case with invasive plant species, the phenology of glossy buckthorn differs from that of other indigenous plant species found in invaded communities. In the forests of eastern Canada, the main phenological difference is a delay in the shedding of glossy buckthorn leaves, which occurs later in the fall than for other indigenous tree species found in that region. Therefore, our objective was to use that phenological characteristic to map the spatial distribution of glossy buckthorn over a portion of southern Québec, Canada, using remote sensing-based approaches. We achieved this by applying a linear temporal unmixing model to a time series of the normalized difference vegetation index (NDVI) derived from Landsat 8 Operational Land Imager (OLI) images to create a map of the probability of the occurrence of glossy buckthorn for the study area. The map resulting from the temporal unmixing model shows an agreement of 69% with field estimates of glossy buckthorn occurrence measured in 121 plots distributed over the study area. Glossy buckthorn mapping accuracy was limited by evergreen species and by the spectral and spatial resolution of the Landsat 8 OLI.

Kernel-Based Nonlinear Spectral Unmixing with Dictionary Pruning

Spectral unmixing extracts subpixel information by decomposing observed pixel spectra into a collection of constituent spectra signatures and their associated fractions. Considering the restriction of linear unmixing model, nonlinear unmixing algorithms find their applications in complex scenes. Kernel-based algorithms serve as important candidates for nonlinear unmixing as they do not require specific model assumption and have moderate computational complexity. In this paper we focus on the linear mixture and nonlinear fluctuation model. We propose a two-step kernel-based unmixing algorithm to address the case where a large spectral library is used as the candidate endmembers or the sparse mixture case. The sparsity-inducing regularization is introduced to perform the endmember selection and the candidate library is then pruned to provide more accurate results. Experimental results with synthetic and real data, particularly those laboratory-created labeled, show the effectiveness of the proposed algorithm compared with state-of-art methods.

Estimation of phytoplankton taxonomic groups in the Arctic Ocean using phytoplankton absorption properties: implication for ocean-color remote sensing.

Knowledge on the phenology and distribution of phytoplankton taxonomic groups (PTGs) represent valuable information when studying marine ecosystem, especially in the Arctic Ocean where rapid warming has drastic effects on sea-ice dynamics, which affect the marine food web. Taxonomic groups of phytoplankton can be discriminated based on their pigment signatures, which, in turn, impact their absorption spectra, given that different pigments have different absorption windows in the visible. Using concurrent measurements of phytoplankton diagnostic pigments and absorption spectra (aph) collected in the Bering and Chukchi Seas, a novel and direct approach was designed for simultaneously estimating the biomass concentrations of several PTGs (Ci) as well as their specific absorption coefficient. The chemotaxonomic tool CHEMTAX was applied to twelve diagnostic pigments measured by high-performance liquid chromatography (HPLC). Their results revealed that the phytoplankton community composition was made of nine groups, from which six dominant were identified: diatoms, dinoflagellates, c3-flagellate, haptophytes type 7, two types of prasinophytes. Out of 117 samples, twenty pairs of Ci derived by CHEMTAX and measured aph were randomly selected and used in a linear unmixing model to extract the specific absorption spectral of each group. This step was repeated 1000 times to provide the mean specific absorption of a given phytoplankton group. These specific absorption spectra were used to reconstruct total aph, which was consistent with the measured aph (R2 from 0.8 to 0.95) at all visible wavelengths (400-700 nm). The derived specific absorption spectra were further used with the measured aph(λ) at ten Moderate Resolution Imaging Spectroradiometer (MODIS) wavebands in a linear unmixing model to test the ability to retrieve the concentrations of PTGs from satellite remote sensing. A comparison between estimated and measured Ci showed that the approach used in this study performed best when retrieving five groups (i.e., dinoflagellates, c3-flagellate, haptophytes, two types of prasinophytes) from the nine initially identified using CHEMTAX with a mean absolute percentage error (MAPE) <35%, except for diatoms with a MAPE value of about 45%. Our approach provides a practical basis for estimation of PTGs using aph(λ) derived from satellite observations and field measurements.

Unmixing Polarimetric Radar Images Based on Land Cover Type Identified by Higher Resolution Optical Data Before Target Decomposition: Application to Forest and Bare Soil

Extracting information from a polarimetric radar representation usually consists in decomposing it with target decomposition algorithms. This first step can be seen as a geometric analysis of the polarimetric information: the identification of physical radar scattering mechanisms. The problem is that average physical parameters are estimated. As a consequence, these parameters might not describe correctly any of the land cover types that can be mixed together into the radar resolution cell. Therefore, using the polarimetric parameters for land cover classification is challenging. The novelty of the method is to propose a thematic analysis of the polarimetric information preceding the geometric one. The objective is to assess if splitting off polarimetric information on a land cover type basis before applying usual target decomposition algorithms can produce more consistent radar scattering mechanisms when land cover classes are mixed inside the radar resolution cell. A cooperative fusion framework in which very high-resolution optical images are used to unmix physical radar scattering mechanisms is proposed. For bare soil and forests, we point out that a linear unmixing model applied to the covariance matrix is able to split off polarimetric information on a land cover type basis. The assessment of the unmixed radar matrices is carried out with polarimetric radar images from the Radarsat-2 satellite. It was found that despite speckle, the reconstructed radar information after the unmixing process is statistically relevant with the observations. The question whether the unmixed radar images contain relevant thematic information is more challenging, but results tend to validate this property. This method could be used to have a better estimation of vegetation biomass in the context of open forested areas.

Group Low-Rank Nonnegative Matrix Factorization With Semantic Regularizer for Hyperspectral Unmixing

In this paper, the low rank prior of abundances of hyperspectral data is explored and combined with semantic information to develop a new Group Low-rank constrained Nonnegative Matrix Factorization (GLrNMF) method for linear hyperspectral unmixing. First, hyperspectral image pixels are divided into several groups of superpixels, and then low-rank constraints are cast on them to explore the semantic geometry in both spatial and spectral domains. By incorporating semantic information into the NMF, we can recover more accurate endmembers and abundances in the linear unmixing model. Some experiments are taken on several synthetic and real hyperspectral data to investigate the performance of GLrNMF, and the results show that it can outperform some state-of-the-art unmixing results.

Predicting temperate forest stand types using only structural profiles from discrete return airborne lidar

Light detection and ranging (lidar) data have been increasingly used for forest classification due to its ability to penetrate the forest canopy and provide detail about the structure of the lower strata. In this study we demonstrate forest classification approaches using airborne lidar data as inputs to random forest and linear unmixing classification algorithms. Our results demonstrated that both random forest and linear unmixing models identified a distribution of rainforest and eucalypt stands that was comparable to existing ecological vegetation class (EVC) maps based primarily on manual interpretation of high resolution aerial imagery. Rainforest stands were also identified in the region that have not previously been identified in the EVC maps. The transition between stand types was better characterised by the random forest modelling approach. In contrast, the linear unmixing model placed greater emphasis on field plots selected as endmembers which may not have captured the variability in stand structure within a single stand type. The random forest model had the highest overall accuracy (84%) and Cohen’s kappa coefficient (0.62). However, the classification accuracy was only marginally better than linear unmixing. The random forest model was applied to a region in the Central Highlands of south-eastern Australia to produce maps of stand type probability, including areas of transition (the ‘ecotone’) between rainforest and eucalypt forest. The resulting map provided a detailed delineation of forest classes, which specifically recognised the coalescing of stand types at the landscape scale. This represents a key step towards mapping the structural and spatial complexity of these ecosystems, which is important for both their management and conservation.